Research map / Large language models / Instruction and fine-tuning
Phrase-Grounding-Aware Supervised Fine-Tuning for Chart Recognition via Side-Masked Attention
Koichiro Ito
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SMAM phrase-grounding loss to SFT | Qwen2.5-VL-7B · ChartQA Human split | relaxed accuracy: 76.2 -> 77.1 | Improved |
| add SMAM phrase-grounding loss to SFT | LLaVA-7B · ChartQA Human split | relaxed accuracy: 36.5 -> 38.6 | Improved |
| add SMAM phrase-grounding loss to SFT | ChartInstruct-7B · ChartQA Human split | relaxed accuracy: 34.5 -> 37.0 | Improved |
| add SMAM phrase-grounding loss to SFT | Qwen2.5-VL-7B · ChartQA Augmented split | relaxed accuracy: 94.9 -> 95.8 | Improved |
| add SMAM phrase-grounding loss to SFT | LLaVA-7B · Chart-to-table average | F1: 72.4 -> 74.1 | Improved |
| add SMAM phrase-grounding loss to SFT with CoT data | Qwen2.5-VL-7B · QA-CoT Human split | relaxed accuracy: 82.5 -> 83.8 | Improved |
| use text localization (bounding-box output) instead of SMAM | LLaVA-7B · Chart-to-table average | F1: 72.4 -> 36.0 (vs SFT) | Hurt |
| use text localization with CoT data | Qwen2.5-VL-3B · QA-CoT Human split | relaxed accuracy: 78.9 -> 75.2 | Hurt |
| use segmentation-based GLaMM-style grounding | LLaVA-7B · Chart-to-table average | F1: 72.4 -> 18.9 | Hurt |
| use text localization on ChartQA | Qwen2.5-VL-7B · ChartQA Human split | relaxed accuracy: 76.2 -> 76.4 | No clear effect |
| add grounding loss L_G alongside L_CE+ (gamma 0.5, alpha 0.1) (ablation) | SMAM on Qwen2.5-VL-7B · ChartQA Human split | relaxed accuracy: 76.8 -> 77.1 | Improved |
| add L_CE+ regularization to L_G (alpha 0.1) (ablation) | SMAM on LLaVA-7B · ChartQA Human split | relaxed accuracy: 37.0 -> 38.6 | Improved |
Closest papers by meaning
- Learning Cross-Modal Context Graph for Visual Grounding (AAAI 2020 · similarity 0.86)
- Grounding of Textual Phrases in Images by Reconstruction (ECCV 2016 · similarity 0.86)
- ChartNet: A Million-Scale, High-Quality Multimodal Dataset for Robust Chart Understanding (CVPR 2026 · similarity 0.85)
- Chart-FR1: Visual Focus-Driven Fine-Grained Reasoning on Dense Charts (CVPR 2026 · similarity 0.85)
- EvoChart: A Benchmark and a Self-Training Approach Towards Real-World Chart Understanding (AAAI 2025 · similarity 0.85)
- Improved Visual Grounding through Self-Consistent Explanations (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
